Enhancing Electoral Equality: Can Education Compensate for Family Background Differences in Voting Participation?

416,946 individuals (Q1 90,922; Q2 101,864; Q3 109,786; Q4 114,374)

All individuals born in Sweden 1970-1974 who completed compulsory schooling at 16 and continued directly to upper secondary school, linked to parents via the Multi-Generation Registry. Mean turnout in the 2010 general election was 89.98%.

Exploits a pilot scheme preceding the early-1990s Swedish upper-secondary reform that lengthened vocational programmes from two years to three. Reform intensity (the share of three-year vocational programmes in an individual's home municipality at age 16) varies across municipalities and cohorts. Models include fixed effects for birth year, home municipality, and both parents' birth years, with standard errors clustered at municipality. The outcome is VALIDATED turnout: the authors scanned and digitised the publicly available Swedish election rolls for the 2010 general election - not survey self-report. Placebo analyses on pre-reform graduating cohorts pass. The scale (417,000 register-validated voters) makes this by a wide margin the best-powered study in the cluster, so its average null is informative rather than merely underpowered.

No average effect of education on turnout, and a large effect concentrated entirely in the poorest quartile. Reduced form for the whole sample: 0.68 percentage points (SE 0.51), not significant. 2SLS for the whole sample: 3.50 points (SE 2.64), not significant. But split by family SES quartile the 2SLS estimates are Q1 +11.62 (4.36, p<0.01), Q2 -0.69 (4.33), Q3 -0.86 (4.79), Q4 -1.47 (13.83). The between-quartile differences are significant for Q2 vs Q1 (p=0.024), Q3 vs Q1 (p=0.011) and Q4 vs Q1 (p=0.050). The first stage is also concentrated at the bottom: the reform raised the probability of completing three years of post-primary education by 26.4 points for Q1 but only 7.4 points for Q4, so part of the heterogeneity is simply that the reform BOUND on the poorest. The paper's honest headline is that education did not raise average turnout but did narrow the class gap in turnout.

Low. Identification comes from where and when a municipality happened to pilot a longer vocational programme, which no family selected into, plus parental birth-year fixed effects. The outcome is administrative rather than self-reported. This is the design that most directly tests whether the observational education-turnout gradient survives removal of family selection, and on average it does not.

Reproduces the average NULL found by Pelkonen 2012 and Berinsky and Lenz 2011, and by Milligan et al.'s own UK arm. Its distinctive contribution - a large compensatory effect confined to the bottom SES quartile - has not been independently replicated in another country, but it converges with the compensation effect Campbell and Niemi 2016 report for civic KNOWLEDGE among Latinos and immigrants, and with Langton and Jennings 1968.